mirror of
https://github.com/mozilla/gecko-dev.git
synced 2024-10-20 08:45:46 +00:00
0824d18204
We treat segments overlapping as a fatal error, rather than a condition to do nothing, because it happening is usually the result of some bad assumptions on the input ELF, and we don't want to silently ignore those. However, there are cases where a setup /could/ lead to overlapping segments, but would be skipped because elfhack wouldn't be a win anyways. By checking segments overlap later, we allow those to not hard fail. --HG-- extra : rebase_source : deca2051722aeaa959c5e4dae06642908f6d843a |
||
---|---|---|
.. | ||
inject | ||
dummy.c | ||
elf.cpp | ||
elfhack.cpp | ||
elfxx.h | ||
inject.c | ||
Makefile.in | ||
moz.build | ||
README | ||
test-array.c | ||
test-ctors.c | ||
test.c |
Elfhack is a program to optimize ELF binaries for size and cold startup speed. Presently, it is quite experimental, though it works well for the target it was created for: Firefox's libxul.so. Elfhack currently only does one thing: packing dynamic relocations ; which ends up being a quite complex task, that can be summarized this way: - Remove RELATIVE relocations from the .rel.dyn/.rela.dyn section. - Inject a small code able to apply relative relocations "by hand" after the .rel.dyn/.rela.dyn section. - Inject a section containing relocative relocations in a different and more packed format, after the small code. - Register the small code as DT_INIT function. Make the small code call what was initially the DT_INIT function, if there was one. - Remove the hole between the new section containing relative relocations and the following sections, adjusting offsets and base addresses accordingly. - Adjust PT_LOAD entries to fit new offsets, and add an additional PT_LOAD entry when that is necessary to handle the discrepancy between offsets and base addresses, meaning the section offsets may yet again need adjustments. - Adjust various DT_* dynamic tags to fit the new ELF layout. - Adjust section headers. - Adjust ELF headers. See http://glandium.org/blog/?p=1177#relocations for some figures.